Remix clone Hacker News

Law is hard!

In general, the de facto status quo is:

1. For whatever reason*, large swaths of LLM output copy-pasted is easily detectable.

2. If you're restrained, polite, with an accurate signal on this, you can indicate you see this, and won't get downvoted heavily. (ex. I'll post "my internal GPT detector went off, [1-2 sentence clipped version of why I think its wrong even if it wasn't GPT]")

3. People tend to downvote said content, as an ersatz vote.

In general, I don't think there needs to be a blanket ban against it, in the sense of I have absolutely no problem with LLM output per se, just lazy invocation of it, ex. large entry-level arguments that were copy-pasted.

i.e. I've used an LLM to sharpen my already-written rushed poor example, which didn't result in low-perplexity, standard-essay-formatted, content.

Additionally, IMHO it's not bad, per se, if someone invests in replying to an LLM. The fact they are replying indicates its an argument worth furthering with their own contribution.

* a strong indicator that a fundamental goal other than perplexity minimization may increase perceived quality

▲

og_kalu 2 days ago | parent | next [-]

The reason is not strange or unknown. The text completion GPT-3 from 2020 often sounds more natural than 4. The reason is the post training processes. Models are more or less being trained to sound like that during RLHF. Stilted, robotic, like a good little assistant. Open AI, Anthropic have said as much. It's not a limitation of the loss function or even state of the art.

▲

refulgentis 2 days ago | parent [-]

I can't give in to misguided pessimism - "Open AI, Anthropic have said as much" is especially not something I can support!

I'm hearing some of the ideas on my corner of llm x creativity Twitter expressed clunkily and if its some irrevocable thing.

You're right the default is to speak like an assistant.

You're wrong that its forced and immutable and a consequence of RLHF and the companies say its so. https://x.com/jpohhhh/status/1784077479730090346

You're especially wrong that RLHF is undesirable https://x.com/jpohhhh/status/1819549737835528555 https://x.com/jpohhhh/status/1819550145522160044.

It's also nigh-trivial to get the completion model back https://x.com/jpohhhh/status/1776434608403325331

I don't know when I'll stop seeing surface-level opinions disguised as cold technological claims on this subject. I would have thought, by now, people doing that would wonder why the wide open lane hasn't been taken, at least once.

	▲	og_kalu 2 days ago \| parent [-]
		I don't understand what you're getting at here. No idea why you've put tweets from a random? person to make your point. Yes these guys have all noted on the effects of post-training on the models. "We want people to know that they’re interacting with a language model and not a person." This is literally a goal of post-training for all these companies. Even when they are training it to have a character, it mustn't sound like a person. It's no surprise they don't sound as natural as their base counterparts. https://www.anthropic.com/research/claude-character >You're wrong that its forced and immutable and a consequence of RLHF and the companies say its so. I never said it was immutable. I said it was a consequence of post-training and it is. All the base models speak more naturally with much less effort. >You're especially wrong that RLHF is undesirable I don't understand what point you're trying to make here. I didn't say it was undesirable. I said it was heavily affecting how natural the models sounded. >It's also nigh-trivial to get the completion model back Try getting GPT-4o to write a story with villains that doesn't end with everyone singing Kumbaya and you'll see how much post-training affects the outputs of these models.

▲

aspenmayer 2 days ago | parent | prev | next [-]

To me, the essence of online discussion boards is a mutual exchange of ideas, thoughts, and opinions via a shared context, all in service of a common goal of a meeting of minds. When one party uses LLMs, it undermines the unspoken agreement to post “authentic” content as opposed to “unauthentic” content. Authenticity in this context is not just a “nice to have,” but is part and parcel to the entire enterprise of participating in a shared experience and understanding via knowledge transfer and cross-cultural exchange.

I can see that you care enough to comment here in a “genuine” and good faith manner, as I recognize your username and your posting output as being in good faith. That being said, an increase in LLM-generated content on HN generally is likely to result in an associated increase in the number of bad actors using LLMs to advance their own ends. I don’t want to give bad actors any quarter, whether that be wiggle room or excuses about Guidelines or on-topic-ness, or any other justification for why self-proclaimed “good” actors think that using LLMs is okay when they do it, but not when bad actors do it, because doing so gives cover to bad actors to do so, as long as they don’t get caught.

	▲	refulgentis 2 days ago \| parent [-]
		> That being said, an increase in LLM-generated content on HN generally is likely to result in an associated increase in the number of bad actors using LLMs to advance their own ends. This hit me like a ton of bricks, very true. The older I get the more I understand the optimist in me rushes to volunteer good things that'll happen over the obvious bad. This, in retrospect, will apply here too and is explanatory for some notably bad vibes I've had here the past year or two. (been here 15 years)

▲

vunderba 2 days ago | parent | prev [-]

Additionally, IMHO it's not bad, per se, if someone invests in replying to an LLM. The fact they are replying indicates its an argument worth furthering with their own contribution

And once those floodgates are open, what exactly makes you think that they're not just also using an LLM to generate their "contribution"?

	▲	refulgentis 2 days ago \| parent [-]
		Not necessarily bad either! Thats what the downvote button is for :)